US Copyright Office Denies Protection for Another AI-Created Image – Slashdot

Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!
Nickname:
Password:
Nickname:
Password:
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
To bolster their claims, artists will use AI to inspire them, but add sufficient non-AI content to ensure they won’t get rejected by the copyright office.
Good luck getting them to publish the non-copyright-able AI-generated inspirations.
Alternatively, just don’t list the tools you use to make your art as if it were a creator.

Alternatively, just don’t list the tools you use to make your art as if it were a creator.
This will work only if you don’t get caught. Otherwise, it would be considered fraud, tampering with a government document i.e. the application for copyright registration, etc. etc.
Granted, the odds of getting caught are low, at least for now.
BUT… you don’t know if a family member or co-worker or collaborator who “knows your secret” may turn on you years later.
We also don’t know if a future technology will be able to “suss out” at least some things that were 100% AI-generated.
Bottom line: Your suggestio
“This will work only if you don’t get caught. Otherwise, it would be considered fraud, tampering with a government document i.e. the application for copyright registration, etc. etc.”
It is no more fraud than not listing a brush and paint when filing the copyright on a painting. The AI does not add any sort of creative element, therefore it is not a creator and should not be listed as one.
“Bottom line: Your suggestion might work, but it is not without legal risks.”
The rest of your post erroneously applies so
“or co-worker or collaborator”
I thought with return to work we were supposed to be doing more collaboration.
I think this is a very flawed finding that will be ultimately overturned (if challenged). What is the difference between taking a photograph of something completely natural (and therefore the photographer only observed) and a Midjourney produced image where the artist actually directs a computer to produce an image.
If photos are copyright-able, then so should Midjourney created images.
Photographs are composed by the photographer. They make choices about framing, lighting, focus, where and when to shoot.
This requires talent, expertise, and effort.
You can’t say the same for telling a computer to make a picture.
Yes you absolutely can say the same about using a computer to do the work. The fact that someone else might have created the code and trained the AI does not change the fact that someone competent had to use the tool to create the art. He didnâ(TM)t just grab an image that the computer spat out. He composed it, using commands to Midjourney.
No they reduce to the same thing, if you are simply telling the computer to make a picture you are simply telling the handheld computer to record a picture.
After all you are telling the computer what subject to generate, what style to apply, what kind of lighting/effects then tweaking those choices and finally selecting which of the many images generated fits your vision or you best like.
BUT… in both cases you need to remember it is a derivative of existing work, just like a recording has the baggage of t
It’s not a derivative work.
If it were, it would have been eligible for copyright protection as one.
The standard for determining what is a derivative work is whether it contains sufficient recognizable copied elements.
These tools are not copying elements. They are recreating them.
You could not be further off the mark with the aid of an electrified mark-missing machine.
“These tools are not copying elements. They are recreating them.”
It’s a mechanical system therefore it is not recreating but copying, there is no sentience and therefore no creative element in the operation of the machine. The machine produces nothing that didn’t come from the input. If I took the mona lisa and applied a twirl effect in photoshop it would still be a derivative work, the only reason I wouldn’t need Leo’s permission to copy and distribute it is because that work is in the public domain.
“If it
If you used a Photoshop effect you would be transforming the actual image. LLMs don’t store the images. Once they are trained on them they are discarded from the software’s point of view. Your interpretation depends on a misunderstanding.
“LLMs don’t store the images. Once they are trained on them they are discarded from the software’s point of view.”
False. The data is stored in the weights and is regurgitated in the output. Your interpretation depends on a misunderstanding.
The data can’t be stored in the weights, because the weights are not large enough. There is less than one byte per training image in the large models we use today. It’s literally impossible to store the images in the model.
“Yes it does, from the start it drew the “avocado chair”, something not in the training set.”
The combination might be semi-novel but every piece of it was contained in the training set +/- any alterations from training feedback [actually just another input] and the mechanical algorithms within the model [another form of input]. The idea to mix data from avocados and chairs might be novel but that came from the operator NOT the tool.
“computes”
Input data
“gradients”
Input data
“So it combines knowledge from mult
I can take the current value out of a prng and the seed and restore the 2000 PB of data generated between its start and now despite those values being less than 1k combined.
I wonder if you realize how disingenuous your statement is. You know for a fact that generative AI models do not store enough relational data to reproduce training images perfectly, or even anywhere close to perfectly. There are quadrillions of differences between any training image compared to almost any image generated by an image-ba
You just fed his foolish fire there with the word compress. The model in no way contains a compressed representation of the image. It only contains information on similarities between images, NONE of which are stored in the model.
That’s not quite correct.
Knowing that a bunch of images are ‘similar’ does not let you produce an image similar to those images, without any of the original images.
You *must* have stored some representation of information drawn from the images themselves as well.

You can’t say the same for telling a computer to make a picture.
Not for a single picture, but I know serious, er, “AI designers” who take a day or more to craft the “perfect” AI-generated image.
The full process involves testing dozens of variations of the input, then once the system begins generating images close to what the person wants, generating hundreds of them, evaluating one by one, selecting the best ones, providing them back to the system, crafting new inputs to generate variations atop those, in several distinct batches using distinct specific values, several
> Photographs are composed by the photographer. They make choices about framing, lighting, focus, where and when to shoot. This requires talent, expertise, and effort. You can’t say the same for telling a computer to make a picture.
What you said reads like someone told a photographer “all you have to do is push a button on the camera, that’s not creative!” Now imagine explaining why that’s wrong to someone who doesn’t even know what the f-stop is.
Because there actually is a lot of skill that goes into c
At a cost of $65 per application, no one is spamming the copyright office with millions of AI generated works of art!
“You can’t say the same for telling a computer to make a picture.”
Tell us you’ve never tried without telling us.
You can get surprisingly good results with just a single prompt, but most of the AI generated images you can’t immediately tell are such were the result of over a dozen prompts, inpainting and outpainting, and so on.
Even the crafting of a single prompt may take significant time and multiple attempts, and should qualify as creative effort in its own right. Practically any snapshot qualifies for cop
“You can’t say the same for telling a computer to make a picture.”
Isn’t that what your doing when you press the button on a camera?
One moment there isn’t a picture, the next you press a button on a camera, a keyboard, or elsewhere and a picture now exists due to your effort. Sounds like its your’s to me.
The photographer decided what they wanted the picture to look like. They saw the image before pressing the button.
The AI-generated image came as a complete surprise to the person who typed in the prompt. They had seen none of it previously.
You’ve obviously not spent even a superficial amount of time knowing how serious users of generative AI actually use it.
Who owns the copyright for something AI generated? The dude that gave the system the input? The team that created the algorithms and code driving the AI? All the authors who came before whose material was ingested (frequently without consent) into the AI to feed its generative model?
A photograph is easy, I took it, I own the copyright on it. I’m the one who saw something photo worthy, configured my camera, framed the photo, and possibly touched it up….

Who owns the copyright for something AI generated? […] The team that created the algorithms and code driving the AI?
Isn’t that a bit like asking whether or not the creator of the camera has any claim over the image that was created by it?
Were people asking these same questions when the first cameras were invented?
I think that the person to gave the prompt to the computer owns the product
“A photograph is easy, I took it, I own the copyright on it.”
That depends on WHAT you took a picture of. If I paid you to take that photo then it is a “work for hire” and I own the copyright, not you. If you took a picture of my painting then you do own the copyright but it is also a derivative encumbered by my copyright and the former means you’ll need permission to exercise the rights of the latter. This is no different than if I write a song and you do a performance, both the performance and the song hav

I also think exercising the copyright [distribution of copies] requires permission from those who own the copyright on the input.
That seems virtually impossible for Generative AI. Generative AIs are trained on countless works, frequently without the permission from those that own the copyright, but if they had it, how do you do derivative licensing when there are potentially millions of authors you’d need to pay royalties to?
Here’s a fun one, tell ChatGPT to write a “day in the life of” story for Spot on the Enterprise. You’ve got major problems if this is anything other than an exercise in amusement you never intend to monetize o

That’s a pretty basic idea you’d never be able to copyright.
Why? If a human wrote that it would be copyrightable.
A story about a cat named Spot, on a ship named the Enterprise, constructed by an organization called the United Federation of Planets is copyrightable. The character (Spot) and settings (Enterprise and UFP) are also copyrightable. Outside of an exception, like parody, you can’t use them for your own creative work without permission.
The concept of a cat on a spaceship is most definitely not copyrightable. Paramount can’t sue you if you write a story about a cat you made up, on a spaceship you made u
I think I misunderstood your comment. Yes, the *idea* isn’t copyrightable, but the story that comes out is. It is copyrightable because in the same way that the output of other algorithms is copyrightable. If I draw a picture, then run a filter over it, the resulting output is copyrightable. The algorithm transformed the input, but the copyright remains.
The only thing that makes AI interesting is that it was trained on a broad selection of already-copyrighted works. So the input is copyrighted by other

Yes, the *idea* isn’t copyrightable, but the story that comes out is.
If it’s created by a human being. 🙂

If I draw a picture, then run a filter over it, the resulting output is copyrightable.
You drew the picture though. Not the same thing. The written analogy would be using AI to proof read, grammar check, and edit your work. You don’t lose copyright by running your novel through a word processor. You still wrote the novel. It’s different than giving ChatGPT a starting premise and trying to claim copyright over the output. You didn’t create a darn thing and “cat on a spaceship” is not a unique or copyrightable idea, so what exactly is here to copyright
>If I paid you to take that photo then it is a “work for hire” and I own the copyright, not you.
Not unless it’s in the contract like that. And then you pay the professional photographer usually at least triple their usual going rate so you retain copyright instead of having a perpetual license to use the images in restricted cases. Don’t like that deal? Then go buy a DSLR and do it yourself. Then you WILL have the copyright.
>The digital image contains unique expression and I think the copyright office
“Not unless it’s in the contract like that.”
Its always about what the contract says, IF there is one. It could even say the owner is a 3rd party.
And its moot about paying triple the rate. Not even sure why you brought that up. You may also have to pay for the photographer’s lunch if that was agreed on. Should we discuss the menu options now?
No its the same and why copyrights are fundamentally flawed device, if you took a picture of my house, I maintained it, bought it, the team of people created the camera, who built on other peoples work and software you used was created by someone else. You lived in a society that taught you, fed you, gave you inspiration.
Why are you so special that just because you where the last person in the link?
Sure you should get credit for doing the last thing in the chain, and if people want to pay you to do that thi

That’s just like every other industry, from builders, to software engineers, I don’t remember getting a percentage of the sales of the products I develop for 75 years after my death.
It sounds like you don’t own the copyright on any products.
When you create a piece of software and copyright it, then you can license that for sale and negotiate for a percentage of sales for 75 years after your death. Lots of software that works that way. If you create something as a “work for hire” then you don’t own the copyright in that case.
This is consistent for music, photography, paintings, and software.
“It sounds like you don’t own the copyright on any products.”
He/she owns the copyright for the part you quoted.
But, oh no. You own the copyright for the part I quoted. If he were to quote me, we’d be in a recursive copyright loop and the universe may explode. For the love of god, please don’t quote me ewibble. The future of the universe is at stake. Mmm, steak.
Many of the questions being asked don’t really have to do with AI:

Who owns the copyright for something AI generated? The dude that gave the system the input?
Yes. That’s is how we treat it with other tools and algorithms. It applies to things like: a crayon, a Spirograph, Photoshop, or C# code. AI should be no different. ***

The team that created the algorithms and code driving the AI?
No, copyright is not assigned to the creator of the tool unless the tool was also created by the dude that gave the system the input. Like Machinima [wikipedia.org] or the demo scene [wikipedia.org].

All the authors who came before whose material was ingested (frequently without consent) into the AI to feed its generative model? ***
THAT is the key question here. Who gave the system the input, when the system scrapes the entire internet?
“A photograph is easy, I took it, I own the copyright on it.”
Did you invent that camera? Did you create it? Did you decide how it handles filtering and the other 8 million options it does internally? Did you build the camera? Do you even understand how it works?
One doesn’t need to be able to understand the manufacturing process for ink and paint, or how they evaporate at a molecular level, or how they bind to paper and canvas, before using them to express a creative impulse. Nobody would claim with a straight face you should have to describe any of those processes before you can copyright a painting.
And if you take a picture of an uncopyrightable AI-generated image?

Suppose another “artist” used a another AI program and came up with the exact same work of “art”. Would that be copyright infringement?
Yes. If I write a poem, and you independently created the exact same poem afterwards, you’ve violated my copyright.

Two different “artists” using different AI programs, creating the same product. Would the first one be more valid than the second? Would the second be less valuable than the first?
Yes and yes. Whoever did it first must’ve done so through their own creativity. The second must’ve copied the first, whether that’s by paint brush strokes or by generative AI input.
By the way, the same art being generated by accident is nearly nonexistent. Even if the text prompt and all the settings were exactly the same (already unlikely), there’s a 1 in 4 billion chance of an exact duplicate
Actually no it would not be an infringement from https://www.legalmatch.com/law… [legalmatch.com]:

A copyright infringement action is unlikely to succeed if the defendant can demonstrate that the works were developed independently of one another. They would have to demonstrate that they did not have access to the copyrighted work and had no way of becoming aware of it prior to creating their own works.
So long as you developed it independently and can prove it you are safe.
I am not sure that being protected from the claim of copyright infringement is the same thing. The law still favours whoever did it first.
Copyright is a product of the law of the land. It specifically protects the ability to use, claim credit, or generate commercial value over a form of human expression to the human that made that expression.
Law is a product of a society of humans, and humans only. The law only protects the rights of humans to operate in human society. AI is not human, and has no rights, privileges, or protections under human law because it’s not human.
The courts have alre
The human expression is precisely the prompts given to the AI to produce the image. The AI did not just produce an image by itself that the copyright filer is then claiming. He directed it – very precisely too 624 text prompts in all. There is also almost no way to replicate what he has done without specific application of his prompts and other edits. This is a creative work – no question. It should be entitled to copyright protection in exactly the same way as a photograph.

This is a creative work – no question. It should be entitled to copyright protection in exactly the same way as a photograph.
In fact, there is a question. In your opinion there is no question. But while you are entitled to your opinion, your opinion has no bearing on the law. And caselaw has already settled that that simply isn’t enough. The Monkey selfie case is a good example. Slater did all the framing, he did all the lighting, he set everything up, but the monkey snapped the photo. Right there the monkey initiated creation, and the monkey has no legal rights as a non-human and therefore cannot assign rights to a human if they never existed.
624 text prompts are less creative, in my opinion, than what Slater did with the Monkey selfie. But my opinion also has no bearing on caselaw. What settles it is that the AI did the artistic creation, he just set the conditions and trained the AI to do the work, just as Slater did with the monkeys. But the AI did the creation, and as the AI has no rights under the law, it never created a copyright protection and therefore could not assign copyright protections.
The question is legal process and the rights of non-human actors under the law, of which there aren’t any.
Just to put some maths on it, if we imagine the 624 prompts to be binary (they clearly arenâ(TM)t) then what we would have is 2^624 possibilities which is approximately 7*10^187 possibilities. To put that number in context – that is more possibilities than there are atoms in the known universe (up to approx 10^82), and represents a lower bound on the number of possibilities. This person has used Midjourney to produce exactly one image that cannot be reproduced by chance – it is impossible (almost surel
However, it’s better defined as the Threshold for Originality. In
I donâ(TM)t think the Wikipedia article that you linked to says what you possibly think it does.
It is quite clear, for the USA at least, that provided that there is direction by a human, then a work can be considered to be copyrightable.
The copyright office is grappling with the issue of tools such as Midjourney making creation easier. However, in much the same way that photographs can be and are copyrighted, so will works created using AI.
A camera is not a human yet the camera is the thing that actually rendered the image. Yet the human gets all the credit.
The “AI” in this case is equivalent to a camera.
Unless you painted the picture, you didn’t actually create it, the camera did. What the photographer did was to make all the choices about lighting, framing and a hundred other things.
At some point prompt writing will be recognized as an art form just like photography is now.
1) Baker v. Selden [justia.com] set that systems cannot be copyrighted, but exact expressions can.
2) Burrow-Giles Lithographic Company v. Sarony [justia.com] set that the human photographer chooses the framing, the angle, the lighting, all to create an image that expresses a specific emotive response. Thus the camera does not confer copyright, but the human choice of details create an image that expresses a human thought.
AI is a system. It cannot be copyrighted, which is a weak but somewhat relevant topic. But more importantly, the Sarony case set the idea that you can take two pictures of the exact same thing from different angles and framing and evoke a very different emotive response to the picture, thus the choice of framing and details are actually human input and copyrightable. AI does this work for the human. It is not copyrightable.
You seem to be proving my point.
The camera and the AI are equivalent in that they are just a system or tool that the human operates.
My point is that the human should be the owner of the copyright. The fact that it was made by an AI is irrelevant (in my opinion).

What is the difference between taking a photograph of something completely natural (and therefore the photographer only observed) and a Midjourney produced image where the artist actually directs a computer to produce an image.
The difference for the copyrightable photo is that a human took the photo after determining the subject matter, lighting, time, temperature, conditions and composition created a work of art.

Also, ask the Dave Slater, a wildlife photographer who’s camera was briefly taken by a bonobo which then snapped a selfie of itself grinning into the camera and was denied copyright protection of the work. Or ask the elephant owners who tried to copyright the artwork that the elephants painted…. The copyright bills are all clearly worded, that it needs to be created by a human. You can contend that especially in the case of the elephant where it paints a picture at the request of the human handlers is no different than an AI algorithm doing the same thing at the request of the human. If the human made it him/herself, it would be copyrightable, but if they didn’t, it isn’t.

Sure you can go on and on about using tools and technology to help develop new art or generate a better result (such as the differences that something like an airbrush allowed vs traditional paint or using photo editors to remove red eye, or make people look “more fit”, etc…). The blurry line seems to be crossed when the human isn’t the one directly creating the work but instead letting an algorithm do it for them. This is the same issue as has been discussed in book authorship and what is and isn’t copyrightable (i.e. fact data is not copyrightable, numbers are not copyrightable, math is not copyrightable). As a result a mathematical algorithm creating an output, that output is not copyrightable.
All this is, is people wanting an AI to generate thousands of images, have them own the copyright to them all, copyright anything that looks close to it generated by another AI, work very little and force everyone to pay royalties for something that generated images after being fed copyrighted works.
They just want a decopyrighter, it just takes existing works, comes up with something unique out of it, and calls it their own. Why don’t you base your art using an AI strictly trained on art and images etc, tha
He asked midjourney to create this art. Midjourney is just a tool. No different than splattering paint on a canvas. You didn’t make the canvas, you didn’t make the paint colors, and in fact you even if you splashed paint on the canvas nature and hydrodynamics determined the end result.
Yet the copyright office will grant you rights to it. AI generated art is human generated.
He didn’t know what the picture was going to look like at any moment before the computer output it for him.
He didn’t make any of those choices. He just described an idea, which is not copyrightable matter.
By that logic, splattering paint on canvas is also not copyrightable either. The person doing the splattering also has no idea what it would look like at the end.
Ok.. so then the AI artist is an experimentalist. Doesn’t mean he can’t copyright the product of his idea. An AI artist is no different than Jackson Pollock, and probably most other artists as well. Do you know any novelists who had their entire book in their head before they started writing? Also, how many books don’t borrow plot ideas and character concepts from other books or people?
If a photographer knew what they were getting, they would only take one picture of any given thing.
Have you ever known a photographer who didn’t carry around a lot of extra storage for all the extra photos they take? Why? Because they know that only a few of the hundreds of photos they take are actually going to turn out the way they want.
Doesn’t the same apply to Jackson Pollock splashes of paint on canvas? The general form is known, but the final result has randomness and natural process running all through it.
If the same prompts were provided a second time, would the image be the same? If it is unique then it is created by the AI; if the same image is generated then it is created by the human.
Agreed but with the understanding that in this case the paint is existing copyrighted work much like a performance of an already recorded song, yes the performance gets a unique copyright but it is encumbered by a requirement to respect the copyright on the song lyrics.
Ah, but you chose which paints to use and what order.
Not really, you’re saying if I reached into a box with paints in it and randomly threw whatever I grabbed blindfolded onto the canvas I can’t copyright the resulting art? Since when has foreknowledge been a requirement for copyright-ability. I bet a lot of artists have created things by accident.
No, what I was saying is that your characterization of paint-splatter as being entirely without human input was incomplete. I’m not sure yet whether a work that was generated using an ai-powered tool should qualify for copyright; on the one hand, I feel like it’s not enough human effort to deserve it, but on the other I feel like it’s similar to a photographer who takes photos in bursts of a dozen and then picks out the one that looks good…
But even if an AI-generated image isn’t itself copyrightable, the

“This is going to create new and creative problems for the copyright office in ways we can’t even speculate yet.”
Well now…leave it to the ‘damn’ machine to show us humans just how broken the entire concept of copyright, truly is. Wouldn’t be the first time humans have been abhorrently guilty of ignorance driven by Greed. Part of being human at this point.
CopyWrong Victim; You’ve been bitching about a way to dismantle the current system of perpetual Greed within copyright for quite some time…why argue about the form of your messiah?
Roll with it. Who knows, the hypocr..er I mean humans might end up accepting wha

Who knows, the hypocr..er I mean humans might end up accepting what they created, was intelligence after all.
Except there’s nothing there, just large matrices of data. It doesn’t think, know, or understand.
How do you know humans do? Our brains are just a large collection of neurons with sensory input, how can you be sure we can think and understand and can’t be represented by a large enough matrix of data. All we do is judge from the output, we have no idea what it means to be intelligent in terms of process.
“intelligence”
Sadly, it isn’t though it has no sentience whatsoever, not even emotionless sentience.

“intelligence”
Sadly, it isn’t though it has no sentience whatsoever, not even emotionless sentience.
I promise that humans are stupid enough to invoke everything you assume would never happen.
Even if it isn’t intelligent, it will be sold as such. Just like humans do, every damn day.
Could we take denial of copyright protection as a precedent, and start denying it for human-created stuff too? Like bands that claim copyright on riffs? Companies that claim copyright on ideas, rather than expressions of ideas? Just like patents, copyright is being seriously abused, not even mentioning the ridiculous length of the protection.

The office also rejected Allen’s argument that denying copyrights for AI-created material leaves a “void of ownership troubling to creators.
Why is a “void of ownership” any sort of problem? It’s called “public domain”.
Copyright requires a human creator. Any changes to that would require Congress. I will agree that copyright is abused far beyond it’s stated goal “To promote the progress of science and useful arts”.
Define “creator”. Is an artist who uses software to generate images of Mandelbrot fractals allowed to claim copyright? Because in both cases, generative AI and generative fractal software just create images using procedural algorithms.
no wonder they kill all humans
…not more
Stuff created by robots should not be protected by copyright
If you can stand in the same place as someone else, take a picture with your own camera and settings, and get basically the same image, you can still copyright that image as your own. Even if someone else stood there and took a picture as well.
AI on the other hand, you can feed it input, you can’t copyright that like you can’t copyright the location, and the AI will generate the exact same image. Ergo, this one is mine, can’t tell the difference? Too bad.
What does this precedent do for movies created by 2D or 3D computer graphics animation software such as Blender?
People are already starting to use AI to create images and effects [engadget.com] for Blender. Stability AI has created a Blender plugin [stability.ai].
A human scripts it and, at least for now, creates the models then Blender creates the movie but make it easy & a lot of people are going to use AI for a lot of the donkey-work. Surely, using a computer tool to do the donkey work for the creative person is just a natural ad
There may be more comments in this discussion. Without JavaScript enabled, you might want to turn on Classic Discussion System in your preferences instead.
Microsoft To Kill Off Third-Party Printer Drivers in Windows
Microsoft Cuts Ties With the Surface Duo After Just 2 Android Version Updates
The bigger the theory the better.

source